Index Distribution Technique for Efficient Search on Unstructured Peer-to-Peer Networks
Resource indexing is an effective technique for fast, successful_ search on decentralized, unstructured peer-topeer (P2P) networks. An_ index is a summary of resources owned by a node, and is distributed_ over the P2P network; any node having the index can answer queries on_ the location of the resources. While more thoroughly distributed _ indices can make queries answered more quickly with a small hop count,_ in large-scale networks, such a scheme may not be always effective due_ to the large space requirement for keeping indices at each node. We_ propose a new index distribution technique that aims to minimize the_ hop count required for each query by distributing indices over the _ network as uniformly as possible, but still in a space-efficient way._ To do so, we compute the weight of each index that estimates how many_ unique resources each index can locate. We give a large weight to an_ index if it can locate many resources that others cannot. On the other_ hand, if a resource can be located from an index, we decrease the_ weights of other indices that can also locate it. Each node selectively keeps the indices with the largest weights, thus_ increasing the chance of successful queries at the node, while keeping _ the space requirement minimum. Simulation studies show that our_ distribution technique is effective in decreasing hop counts and_ messages needed for resolving queries. It decreases the average hop_ count by up to 44% with 75%-less messages when used with flooding based_ queries. Random-walk with our technique also decrease the average hop_ count by up to 58% with 82%-less messages. Furthermore, the query_ success rate with a limited timeout condition also increases,_ approaching nearly to 100%.
I. INTRODUCTION
Recently, most of the popular peer-to-peer (P2P) networks, e.g., FreeNet [16], Gnutella [14], and FastTrack [15], are unstructured since they can scale up very well along with a high demand of users. In such networks, searching resources such as files is one of the most common and important but complicated tasks. It can take long time and generate a large number of messages occupying the overall network; however, it does not always succeed, especially when searching rare resources in large-scale networks. Resource indexing is one of the approaches to the problems [3, 6]. An index is a summary of resources owned by a node, and is distributed over the P2P network; any node having the index can answer queries on the location of the resources on behalf of the resource owner itself. However, while more thoroughly distributed indices can make queries answered more quickly with a small hop count, in large-scale networks, such a scheme may not be always effective due to the large space requirement for keeping indices at each node. We propose a new index distribution technique that aims to minimize the hop counts of queries by distributing indices over the network as uniformly as possible. We use the Bloom filter [15] to compute the index of a node, which can answer whether a resource is available in the node, but does not always produce correct results. Queries on the existence of a resource succeed with high probability; when it fails, we retry the query to find different nodes. To distribute indices as uniformly as possible, and at the same time in a space-efficient way, we compute the weight of each index that estimates how many unique resources each index can locate. We give a large weight to an index if it can locate many resources that others cannot. On the other hand, if a resource can be located from an index, we decrease the_ weights of other indices that can also locate it. Each node selectively keeps the indices with the largest weights, thus increasing the chance of successful queries at the node, while keeping the space requirement minimum. This proposed index distribution effectively augments with existing query methods for unstructured P2P networks, such flooding [9] and random walk [7], and decrease their average hop counts. Simulation studies show that our distribution technique is effective in decreasing hop counts and messages needed for resolving queries. It decreases the average hop count by up to 44% with 75%-less messages when used with flooding based_ queries. Random walk with our technique also decreases the average hop_ count by up to 58% with 82%-less messages. Furthermore, the query success rate with a limited timeout condition also increases, approaching nearly to 100%.


DOWNLOAD FULL REPORT
http://googleurl?sa=t&source=web&cd=1&ve...bs_all.jsp%3Farnumber%3D4600381&ei=5r4BToyyMKTiiAKEoP2oCA&usg=AFQjCNEa3IU4FnkS29top3Jq5HDAYJZsHQ&sig2=mTrBv3EQyc2d339L6GuNOw